# BEHAVIOR 2026 self-evaluation: 100 tasks

100 tasks, instances 301–310, rollout 0: 1000 episodes. Filenames are `{task}_{instance}_{rollout}`.

## Contents

- `metrics/`: 1000 metrics JSON files.
- `videos/`: 1000 evaluation MP4 files, 31.39 GB total. H264, 30 fps, 672x448.
- `robot/r1pro.yaml`: sha256 prefix `a98fa8a81472adfe`.
- `wrapper/default_wrapper.py`: `omnigibson.eval.wrappers.DefaultWrapper` (224x224 RGB), sha256 prefix `069fbfdc4387c45d`.
- `selection_index.json`: per-episode Q and success.

## Scores

- Episodes: 1000
- Tasks: 100
- Successes: 426
- Macro Q (mean of the 10 instance scores within each task, then the mean across 100 tasks): 0.629446